Retail Pulse Report: AI, Smart Carts, and Singles’ Day
Singles’ Day was last week, but the Chinese consumer wasn’t buying. And confusion still emerges from blurring the lines around the definition of AI - which won't help adoption.
Instead of being darkest before dawn, it might be brightest before the dark for the US economy and its impact on retail, with consumer confidence and retail spending up in the last days and weeks before the election. In AI news, research, companies, and consumers continue to blur the lines between things like computer vision, GenAI, and optimization, which really isn’t going to help with adoption. But one area that seems to be gaining traction – and computer vision plays a large role in making it happen – is smart carts. Let’s dive in!
Retail Economic Indicators
In the US, good-news economic indicators came in way too late for the current administration to take credit. Inflation expectations fell to their lowest level since December 2020, according to the University of Michigan’s consumer sentiment survey. Consumers expect inflation to sit around 2.6% in a year, a decrease from October’s expectation of 2.7%. Consumer sentiment was up in November’s reading, from 71 in October to 73. But the authors note that this read was taken before the election.
The CNBC/NRF Retail Monitor showed strong sales gains in October. Total retail sales were up 4.13% year over year, and core retail sales (excluding restaurants on top of auto dealers and gas stations) were up 4.59%. It’s worth pointing out that this is well above the pace of inflation. The Retail Monitor results tend to be more positive than the US Census Bureau (which hasn’t been released yet), but these are still very, very positive numbers. Clothing and accessories sales were up 8.56% year over year (I think school winter clothes buying finally kicked in), and every category was up, including the big ticket items that have been depressed in part due to high interest rates – only electronics & appliances and sporting goods were down year over year.
The only clouds for the US are on the horizon. Imports are still uncharacteristically high for a time when they normally start to peter out. Why? That strike that fizzled at the beginning of October did so because they agreed to a delay – January 15. Combined with the end of that contract extension, and the threat of impending tariffs, retailers are still pressing on the gas to bring in product early.
Finally, I’m not sure that this is an economic indicator, but it comes from a government, so I’ll put it here: The UK Treasury has said that it will investigate the use of cash in retail, specifically whether retailers are making it too hard for consumers to use cash. They point out that it’s mostly older generations that are left behind when a retailer tries to move them to other forms of payment, and they want to balance the needs of these consumer groups against retailers’ point that accepting cash is increasingly expensive and risky to secure. I’ve seen this at the state level in the US, where some states are explicitly making it illegal to not accept cash. My 83-year-old aunt still writes checks – I had to show her how to use her debit card – so I do get that older generations may particularly struggle. I can also see a future where the POS may be attended, but the cash gets fed into a cash recycling machine so that the store associate handling cash, and the drawer they put it in, become a thing of the past.
AI & Retail
Google Cloud published a whitepaper that was intended to depict 185 real-world GenAI use cases from “the world’s leading organizations”. Yes there were actually 185 of them. But spoiler alert, “leading” does a lot of heavy lifting in their description. And anyway, it’s less interesting which companies are doing what because there is also a lot of heavy lifting in the term “use case” – a lot of these things are POCs or beta tests and have not been scaled. Everything sounds way better than it might actually be.
However, the way they chose to categorize the use cases was interesting. There were six categories, and they all ended in the word “agent” – the new shiny object du jour:
Customer Agents – which are basically chatbots. 45 of them came in here. I actually cynically expected this to be the largest category by far and was a little surprised that there was a lot of room for the other categories.
Employee Agents - help workers be more productive. Oh wait, half of these sounded like customer service chatbots. Half of the use cases listed in this category are companies that offer the service, not the companies that are using them. This got us up to 99, but a lot of them were a stretch. At the end of the day, whether answering employee questions or customer questions – or helping employees answer customer questions – aren’t those all still just… chatbots?
Code Agents - helping developers. This one seemed more genuinely differentiated from chatbots, though I've heard complaints that it takes 3x as long to troubleshoot code that AI wrote than it would be just to write it from scratch. Only 14 options listed here.
Data agents - talking to your data. Another forty or so in here, but again, still just a chatbot. This one is a more interesting use case, because you would theoretically have a wider range of semantics that you would need to train your chatbot on. A customer service chatbot needs to know “where’s my order” but a data chatbot needs to know the full range of the data model and the ways that data might get used in questions. It’s kind of the umbrella type of chatbot for which everything else is more specialized.
Security Agents - pattern recognition in threat assessments. Actually not a chatbot! And there were about twenty here.
Creative Agents - things like writing product descriptions or making 3D images from 2D. This is more actual GenAI (I can’t believe I’m saying it that way). But very interesting to me that only about fifteen use cases made this part of the list.
A survey by Coupon Cabin, conducted in October, asked 1,000 consumers about their knowledge and use of AI in shopping. 96% of respondents said they were at least somewhat familiar with AI, with nearly 82% indicating that the technology is already a part of their lives in some way. I would love to know what definition they were asked about – do we mean GenAI only here? Regardless, 54% said they notice AI will shopping online and about 1/3 find it helpful to online shopping. 60% believe AI is “trustworthy” and nearly 90% believe “AI” will become even more popular over the next few years.
I think some of these results show that companies are being a bit more transparent about when AI is in use and when it is not, which is good. I’ve seen other surveys that said consumers have “never used” AI when clearly they had. Awareness seems higher now. But in general, respondents said they used “AI” for product search and for finding the best deals. Nothing revolutionary there.
And to prove the point that not all AI is GenAI, Amazon wants their drivers to use their Echo Frame smart glasses. I love the title of the article: “Amazon smart glasses could help drivers evade scary dogs”. This is starting to sound more like the Amazon of old. The Echo Frames are consumer-grade tech, but it hasn’t been selling well at all, so not surprising that Amazon might take a look at how it can be used to benefit their own employees. The intent is to help drivers navigate parking lots and buildings to be more efficient. It’s not working great, according to reports – the battery isn’t lasting a full driver shift, and there are a lot of data gaps that the glasses need to collect themselves before it can be of benefit to the drivers. But this is the kind of long play that Amazon used to make. Get it in the field, figure out how to make it better, get on the learning curve, and suck in all available data that might benefit the use case, or Amazon overall. Unless the battery problem simply can’t be solved, I expect we’ll see more about these glasses from them in the future.
Retail Winners and Losers
Hey, did you hear how Singles’ Day went? Me neither. That’s probably because it apparently didn’t go so well this year. As a reminder, Alibaba started the event on 11/11/2009, intended to indulge one’s self before diving into festive seasons that focus on other people. Over time it grew and spread to other platforms, so that it now occupies weeks with multiple sellers offering competing deals and discounts.
Over the last 15 years, it has also grown as an indicator of consumer confidence, and this year the results were not so good. Consumers are saying they no longer trust 11/11 deals – they say they aren’t the lowest prices and also that retailers and brands are raising prices right before the shopping day so that they can claim deeper discounts during the event. Economists and retail experts in China say that the economy has been tough enough that the deals and discounts you would normally see for 11/11 have really already been going on all year, so this shopping holiday is hardly a new incentive for consumers to spend.
Store Innovations
Here’s another thing we haven’t talked about in a long while: smart shopping carts. Instacart made some investments in their Caper Cart, and it looks like they and other competitors may finally be getting some traction. Instacart gained ground with multiple local independent grocers across 8 states in the US. The Caper Carts include cameras, sensors, and a built-in scale. The combination of these technologies mean scanless tracking for consumers – you put it in your cart and it tracks it. It also of course comes with an interactive screen that tracks spending and provides personalized promos and awards and checkout.
Instacart isn’t the only game in town. Intermarche also announced a pilot program with Shopic and Capgemini that will introduce smart carts into a store in Provins, France. This uses a clip-on device on the cart that also leverages computer vision tech to identify items as they are placed in the cart with no scanning required. This one doesn’t sound quite so far along in terms of interactivity or personalization, but will come in later phases.
This is an interesting case of something that has been around a long time before breaking through to any kind of adoption. I remember as early as 2005, companies experimenting with shopping buddies, store-provided scanners, prequels to smart carts. But it seems like scanning was really the big friction point – both for consumers and retailers – and now that computer vision, knit together with other sensors, can overcome that, it’s much more a case of game on for smart carts.
What Did We Learn This Week?
The retail economy seems to be inching its way out of “bad” territory into good. Unfortunately, there are a lot of reasons why that may not last into next year. So enjoy it while it lasts. On the AI front, keep a sharp eye out for mixing metaphors – GenAI is not optimization AI, and chatbots are not computer vision. However, when you dig into it, most of the AI that’s getting adopted with any scale are chatbots.
Singles’ Day isn’t what it used to be, and it’s not clear that it ever will be again. That’s one of the challenges of a shopping holiday that isn’t really anchored in any kind of actual holiday – easy come, easy go. But smart carts seem to have finally found their footing, after nearly 20 years of trying. Not so easy to get here, remains to be seen if it will stay.
If there’s one thing to take away from this week, it’s that past performance in no way predicts future results. Sometimes that means signs of good news before it gets bad. Sometimes it means a lot of struggle before finding a use case or a proof point. Either way, the future remains as murky as ever.